1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
4 */
5
6 #include <linux/fs.h>
7 #include <linux/miscdevice.h>
8 #include <linux/poll.h>
9 #include <linux/dlm.h>
10 #include <linux/dlm_plock.h>
11 #include <linux/slab.h>
12
13 #include "dlm_internal.h"
14 #include "lockspace.h"
15
16 static DEFINE_SPINLOCK(ops_lock);
17 static LIST_HEAD(send_list);
18 static LIST_HEAD(recv_list);
19 static DECLARE_WAIT_QUEUE_HEAD(send_wq);
20 static DECLARE_WAIT_QUEUE_HEAD(recv_wq);
21
22 struct plock_async_data {
23 void *fl;
24 void *file;
25 struct file_lock flc;
26 int (*callback)(struct file_lock *fl, int result);
27 };
28
29 struct plock_op {
30 struct list_head list;
31 int done;
32 /* if lock op got interrupted while waiting dlm_controld reply */
33 bool sigint;
34 struct dlm_plock_info info;
35 /* if set indicates async handling */
36 struct plock_async_data *data;
37 };
38
set_version(struct dlm_plock_info * info)39 static inline void set_version(struct dlm_plock_info *info)
40 {
41 info->version[0] = DLM_PLOCK_VERSION_MAJOR;
42 info->version[1] = DLM_PLOCK_VERSION_MINOR;
43 info->version[2] = DLM_PLOCK_VERSION_PATCH;
44 }
45
check_version(struct dlm_plock_info * info)46 static int check_version(struct dlm_plock_info *info)
47 {
48 if ((DLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
49 (DLM_PLOCK_VERSION_MINOR < info->version[1])) {
50 log_print("plock device version mismatch: "
51 "kernel (%u.%u.%u), user (%u.%u.%u)",
52 DLM_PLOCK_VERSION_MAJOR,
53 DLM_PLOCK_VERSION_MINOR,
54 DLM_PLOCK_VERSION_PATCH,
55 info->version[0],
56 info->version[1],
57 info->version[2]);
58 return -EINVAL;
59 }
60 return 0;
61 }
62
dlm_release_plock_op(struct plock_op * op)63 static void dlm_release_plock_op(struct plock_op *op)
64 {
65 kfree(op->data);
66 kfree(op);
67 }
68
send_op(struct plock_op * op)69 static void send_op(struct plock_op *op)
70 {
71 set_version(&op->info);
72 spin_lock(&ops_lock);
73 list_add_tail(&op->list, &send_list);
74 spin_unlock(&ops_lock);
75 wake_up(&send_wq);
76 }
77
78 /* If a process was killed while waiting for the only plock on a file,
79 locks_remove_posix will not see any lock on the file so it won't
80 send an unlock-close to us to pass on to userspace to clean up the
81 abandoned waiter. So, we have to insert the unlock-close when the
82 lock call is interrupted. */
83
do_unlock_close(const struct dlm_plock_info * info)84 static void do_unlock_close(const struct dlm_plock_info *info)
85 {
86 struct plock_op *op;
87
88 op = kzalloc(sizeof(*op), GFP_NOFS);
89 if (!op)
90 return;
91
92 op->info.optype = DLM_PLOCK_OP_UNLOCK;
93 op->info.pid = info->pid;
94 op->info.fsid = info->fsid;
95 op->info.number = info->number;
96 op->info.start = 0;
97 op->info.end = OFFSET_MAX;
98 op->info.owner = info->owner;
99
100 op->info.flags |= DLM_PLOCK_FL_CLOSE;
101 send_op(op);
102 }
103
dlm_posix_lock(dlm_lockspace_t * lockspace,u64 number,struct file * file,int cmd,struct file_lock * fl)104 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
105 int cmd, struct file_lock *fl)
106 {
107 struct plock_async_data *op_data;
108 struct dlm_ls *ls;
109 struct plock_op *op;
110 int rv;
111
112 ls = dlm_find_lockspace_local(lockspace);
113 if (!ls)
114 return -EINVAL;
115
116 op = kzalloc(sizeof(*op), GFP_NOFS);
117 if (!op) {
118 rv = -ENOMEM;
119 goto out;
120 }
121
122 op->info.optype = DLM_PLOCK_OP_LOCK;
123 op->info.pid = fl->fl_pid;
124 op->info.ex = (fl->fl_type == F_WRLCK);
125 op->info.wait = IS_SETLKW(cmd);
126 op->info.fsid = ls->ls_global_id;
127 op->info.number = number;
128 op->info.start = fl->fl_start;
129 op->info.end = fl->fl_end;
130 /* async handling */
131 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
132 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
133 if (!op_data) {
134 dlm_release_plock_op(op);
135 rv = -ENOMEM;
136 goto out;
137 }
138
139 /* fl_owner is lockd which doesn't distinguish
140 processes on the nfs client */
141 op->info.owner = (__u64) fl->fl_pid;
142 op_data->callback = fl->fl_lmops->lm_grant;
143 locks_init_lock(&op_data->flc);
144 locks_copy_lock(&op_data->flc, fl);
145 op_data->fl = fl;
146 op_data->file = file;
147
148 op->data = op_data;
149
150 send_op(op);
151 rv = FILE_LOCK_DEFERRED;
152 goto out;
153 } else {
154 op->info.owner = (__u64)(long) fl->fl_owner;
155 }
156
157 send_op(op);
158
159 rv = wait_event_interruptible(recv_wq, (op->done != 0));
160 if (rv == -ERESTARTSYS) {
161 spin_lock(&ops_lock);
162 /* recheck under ops_lock if we got a done != 0,
163 * if so this interrupt case should be ignored
164 */
165 if (op->done != 0) {
166 spin_unlock(&ops_lock);
167 goto do_lock_wait;
168 }
169
170 op->sigint = true;
171 spin_unlock(&ops_lock);
172 log_debug(ls, "%s: wait interrupted %x %llx pid %d",
173 __func__, ls->ls_global_id,
174 (unsigned long long)number, op->info.pid);
175 goto out;
176 }
177
178 do_lock_wait:
179
180 WARN_ON(!list_empty(&op->list));
181
182 rv = op->info.rv;
183
184 if (!rv) {
185 if (locks_lock_file_wait(file, fl) < 0)
186 log_error(ls, "dlm_posix_lock: vfs lock error %llx",
187 (unsigned long long)number);
188 }
189
190 dlm_release_plock_op(op);
191 out:
192 dlm_put_lockspace(ls);
193 return rv;
194 }
195 EXPORT_SYMBOL_GPL(dlm_posix_lock);
196
197 /* Returns failure iff a successful lock operation should be canceled */
dlm_plock_callback(struct plock_op * op)198 static int dlm_plock_callback(struct plock_op *op)
199 {
200 struct plock_async_data *op_data = op->data;
201 struct file *file;
202 struct file_lock *fl;
203 struct file_lock *flc;
204 int (*notify)(struct file_lock *fl, int result) = NULL;
205 int rv = 0;
206
207 WARN_ON(!list_empty(&op->list));
208
209 /* check if the following 2 are still valid or make a copy */
210 file = op_data->file;
211 flc = &op_data->flc;
212 fl = op_data->fl;
213 notify = op_data->callback;
214
215 if (op->info.rv) {
216 notify(fl, op->info.rv);
217 goto out;
218 }
219
220 /* got fs lock; bookkeep locally as well: */
221 flc->fl_flags &= ~FL_SLEEP;
222 if (posix_lock_file(file, flc, NULL)) {
223 /*
224 * This can only happen in the case of kmalloc() failure.
225 * The filesystem's own lock is the authoritative lock,
226 * so a failure to get the lock locally is not a disaster.
227 * As long as the fs cannot reliably cancel locks (especially
228 * in a low-memory situation), we're better off ignoring
229 * this failure than trying to recover.
230 */
231 log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
232 (unsigned long long)op->info.number, file, fl);
233 }
234
235 rv = notify(fl, 0);
236 if (rv) {
237 /* XXX: We need to cancel the fs lock here: */
238 log_print("dlm_plock_callback: lock granted after lock request "
239 "failed; dangling lock!\n");
240 goto out;
241 }
242
243 out:
244 dlm_release_plock_op(op);
245 return rv;
246 }
247
dlm_posix_unlock(dlm_lockspace_t * lockspace,u64 number,struct file * file,struct file_lock * fl)248 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
249 struct file_lock *fl)
250 {
251 struct dlm_ls *ls;
252 struct plock_op *op;
253 int rv;
254 unsigned char fl_flags = fl->fl_flags;
255
256 ls = dlm_find_lockspace_local(lockspace);
257 if (!ls)
258 return -EINVAL;
259
260 op = kzalloc(sizeof(*op), GFP_NOFS);
261 if (!op) {
262 rv = -ENOMEM;
263 goto out;
264 }
265
266 /* cause the vfs unlock to return ENOENT if lock is not found */
267 fl->fl_flags |= FL_EXISTS;
268
269 rv = locks_lock_file_wait(file, fl);
270 if (rv == -ENOENT) {
271 rv = 0;
272 goto out_free;
273 }
274 if (rv < 0) {
275 log_error(ls, "dlm_posix_unlock: vfs unlock error %d %llx",
276 rv, (unsigned long long)number);
277 }
278
279 op->info.optype = DLM_PLOCK_OP_UNLOCK;
280 op->info.pid = fl->fl_pid;
281 op->info.fsid = ls->ls_global_id;
282 op->info.number = number;
283 op->info.start = fl->fl_start;
284 op->info.end = fl->fl_end;
285 if (fl->fl_lmops && fl->fl_lmops->lm_grant)
286 op->info.owner = (__u64) fl->fl_pid;
287 else
288 op->info.owner = (__u64)(long) fl->fl_owner;
289
290 if (fl->fl_flags & FL_CLOSE) {
291 op->info.flags |= DLM_PLOCK_FL_CLOSE;
292 send_op(op);
293 rv = 0;
294 goto out;
295 }
296
297 send_op(op);
298 wait_event(recv_wq, (op->done != 0));
299
300 WARN_ON(!list_empty(&op->list));
301
302 rv = op->info.rv;
303
304 if (rv == -ENOENT)
305 rv = 0;
306
307 out_free:
308 dlm_release_plock_op(op);
309 out:
310 dlm_put_lockspace(ls);
311 fl->fl_flags = fl_flags;
312 return rv;
313 }
314 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
315
dlm_posix_get(dlm_lockspace_t * lockspace,u64 number,struct file * file,struct file_lock * fl)316 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
317 struct file_lock *fl)
318 {
319 struct dlm_ls *ls;
320 struct plock_op *op;
321 int rv;
322
323 ls = dlm_find_lockspace_local(lockspace);
324 if (!ls)
325 return -EINVAL;
326
327 op = kzalloc(sizeof(*op), GFP_NOFS);
328 if (!op) {
329 rv = -ENOMEM;
330 goto out;
331 }
332
333 op->info.optype = DLM_PLOCK_OP_GET;
334 op->info.pid = fl->fl_pid;
335 op->info.ex = (fl->fl_type == F_WRLCK);
336 op->info.fsid = ls->ls_global_id;
337 op->info.number = number;
338 op->info.start = fl->fl_start;
339 op->info.end = fl->fl_end;
340 if (fl->fl_lmops && fl->fl_lmops->lm_grant)
341 op->info.owner = (__u64) fl->fl_pid;
342 else
343 op->info.owner = (__u64)(long) fl->fl_owner;
344
345 send_op(op);
346 wait_event(recv_wq, (op->done != 0));
347
348 WARN_ON(!list_empty(&op->list));
349
350 /* info.rv from userspace is 1 for conflict, 0 for no-conflict,
351 -ENOENT if there are no locks on the file */
352
353 rv = op->info.rv;
354
355 fl->fl_type = F_UNLCK;
356 if (rv == -ENOENT)
357 rv = 0;
358 else if (rv > 0) {
359 locks_init_lock(fl);
360 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
361 fl->fl_flags = FL_POSIX;
362 fl->fl_pid = -op->info.pid;
363 fl->fl_start = op->info.start;
364 fl->fl_end = op->info.end;
365 rv = 0;
366 }
367
368 dlm_release_plock_op(op);
369 out:
370 dlm_put_lockspace(ls);
371 return rv;
372 }
373 EXPORT_SYMBOL_GPL(dlm_posix_get);
374
375 /* a read copies out one plock request from the send list */
dev_read(struct file * file,char __user * u,size_t count,loff_t * ppos)376 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
377 loff_t *ppos)
378 {
379 struct dlm_plock_info info;
380 struct plock_op *op = NULL;
381
382 if (count < sizeof(info))
383 return -EINVAL;
384
385 spin_lock(&ops_lock);
386 if (!list_empty(&send_list)) {
387 op = list_first_entry(&send_list, struct plock_op, list);
388 if (op->info.flags & DLM_PLOCK_FL_CLOSE)
389 list_del(&op->list);
390 else
391 list_move(&op->list, &recv_list);
392 memcpy(&info, &op->info, sizeof(info));
393 }
394 spin_unlock(&ops_lock);
395
396 if (!op)
397 return -EAGAIN;
398
399 /* there is no need to get a reply from userspace for unlocks
400 that were generated by the vfs cleaning up for a close
401 (the process did not make an unlock call). */
402
403 if (op->info.flags & DLM_PLOCK_FL_CLOSE)
404 dlm_release_plock_op(op);
405
406 if (copy_to_user(u, &info, sizeof(info)))
407 return -EFAULT;
408 return sizeof(info);
409 }
410
411 /* a write copies in one plock result that should match a plock_op
412 on the recv list */
dev_write(struct file * file,const char __user * u,size_t count,loff_t * ppos)413 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
414 loff_t *ppos)
415 {
416 struct plock_op *op = NULL, *iter;
417 struct dlm_plock_info info;
418 int do_callback = 0;
419
420 if (count != sizeof(info))
421 return -EINVAL;
422
423 if (copy_from_user(&info, u, sizeof(info)))
424 return -EFAULT;
425
426 if (check_version(&info))
427 return -EINVAL;
428
429 spin_lock(&ops_lock);
430 list_for_each_entry(iter, &recv_list, list) {
431 if (iter->info.fsid == info.fsid &&
432 iter->info.number == info.number &&
433 iter->info.owner == info.owner) {
434 if (iter->sigint) {
435 list_del(&iter->list);
436 spin_unlock(&ops_lock);
437
438 pr_debug("%s: sigint cleanup %x %llx pid %d",
439 __func__, iter->info.fsid,
440 (unsigned long long)iter->info.number,
441 iter->info.pid);
442 do_unlock_close(&iter->info);
443 memcpy(&iter->info, &info, sizeof(info));
444 dlm_release_plock_op(iter);
445 return count;
446 }
447 list_del_init(&iter->list);
448 memcpy(&iter->info, &info, sizeof(info));
449 if (iter->data)
450 do_callback = 1;
451 else
452 iter->done = 1;
453 op = iter;
454 break;
455 }
456 }
457 spin_unlock(&ops_lock);
458
459 if (op) {
460 if (do_callback)
461 dlm_plock_callback(op);
462 else
463 wake_up(&recv_wq);
464 } else
465 log_print("%s: no op %x %llx", __func__,
466 info.fsid, (unsigned long long)info.number);
467 return count;
468 }
469
dev_poll(struct file * file,poll_table * wait)470 static __poll_t dev_poll(struct file *file, poll_table *wait)
471 {
472 __poll_t mask = 0;
473
474 poll_wait(file, &send_wq, wait);
475
476 spin_lock(&ops_lock);
477 if (!list_empty(&send_list))
478 mask = EPOLLIN | EPOLLRDNORM;
479 spin_unlock(&ops_lock);
480
481 return mask;
482 }
483
484 static const struct file_operations dev_fops = {
485 .read = dev_read,
486 .write = dev_write,
487 .poll = dev_poll,
488 .owner = THIS_MODULE,
489 .llseek = noop_llseek,
490 };
491
492 static struct miscdevice plock_dev_misc = {
493 .minor = MISC_DYNAMIC_MINOR,
494 .name = DLM_PLOCK_MISC_NAME,
495 .fops = &dev_fops
496 };
497
dlm_plock_init(void)498 int dlm_plock_init(void)
499 {
500 int rv;
501
502 rv = misc_register(&plock_dev_misc);
503 if (rv)
504 log_print("dlm_plock_init: misc_register failed %d", rv);
505 return rv;
506 }
507
dlm_plock_exit(void)508 void dlm_plock_exit(void)
509 {
510 misc_deregister(&plock_dev_misc);
511 }
512
513